Evaluating AlphaFold using Windows 10, WSL and Ubuntu about deepmind-research

[OneWorld-github]

I actually got passed this error doing this...

I did this by uninstalling any version of python other than version 3.6 on my system.

I then uninstalled ubuntu from the Microsoft Store, and reinstalled ubuntu 18.04 from the Microsoft store.

Then run this command in the ubuntu linux command line:

sudo apt-get install python3-venv

I then ran

bash alphafold_casp13/run_eval.sh

which downloads the libraries from internet and installs them.

and got this error message:-

`Collecting grpcio>=1.8.6 (from tensorflow==1.14->-r alphafold_casp13/requirements.txt (line 6))
Downloading 
https://files.pythonhosted.org/packages/e3/0e/f56aa1f8200ae3c5d38305e69f5920caa480c7ff54ae4d8a5f57d1d69fa4/grpcio-1.31.0.tar.gz (20.0MB)  100% |████████████████████████████████| 20.0MB 79kB/s
Complete output from command python setup.py egg_info:
Traceback (most recent call last):
File "<string>", line 1, in <module>
File "/tmp/pip-build-fnozvtho/grpcio/setup.py", line 197, in <module>
   if check_linker_need_libatomic():
File "/tmp/pip-build-fnozvtho/grpcio/setup.py", line 157, in check_linker_need_libatomic stderr=PIPE)
File "/usr/lib/python3.6/subprocess.py", line 729, in __init__ restore_signals, start_new_session)
File "/usr/lib/python3.6/subprocess.py", line 1364, in _execute_child raise child_exception_type(errno_num, err_msg, err_filename)
FileNotFoundError: [Errno 2] No such file or directory: 'cc': 'cc'
----------------------------------------
Command "python setup.py egg_info" failed with error code 1 in /tmp/pip-build-fnozvtho/grpcio/`

[OneWorld-github]

So I tried

sudo apt install python3-pip

in the ubuntu terminal, to see if the package manager there needed an update to get it to work.

and reran this script, in the windows terminal

bash alphafold_casp13/run_eval.sh

and this did indeed work...

`
I0906 17:21:43.850929 140638077060928 contacts.py:81] Writing output to /home/user/contacts_T1019s2_2020_09_06_17_17_42/distogram/2
I0906 17:21:43.902758 140638077060928 contacts.py:114] Eval config is {'crop_shingle_x': 16, 'crop_shingle_y': 16, 'eval_sstable': 'alphafold_casp13/T1019s2/T1019s2.tfrec', 'max_num_examples': 500, 'output_path': '/home/user/contacts_T1019s2_2020_09_06_17_17_42/distogram/2', 'pyramid_weights': 0.75, 'save_rr_files': True, 'stats_file': 'alphafold_casp13/873731/stats_train_s35.json'}
num_bins: 64
I0906 17:21:45.237473 139822654359360 contacts.py:191] SepWorking on 0 T1019s2-l64_s24 T1019s2 64
I0906 17:21:45.276877 139822654359360 contacts.py:194] Getting residue_index from features
I0906 17:21:45.674456 140638077060928 contacts.py:191] SepWorking on 0 T1019s2-l64_s24 T1019s2 64
I0906 17:21:45.773195 140638077060928 contacts.py:194] Getting residue_index from features

`

hmm how do I access

/home/user/contacts_T1019s2_2020_09_06_17_17_42/distogram/2

from windows, if it is running Ubuntu through WSL ?

[OneWorld-github]

My system has a single NVidia 1070 GTX GPU (8GB ram), and it is running Alpha Fold seemingly slowly, and seems to freeze other operations in Windows. Wondering if this is because I'm not using a fast enough GPU or whether this Windows, Ubuntu, WSL setup introduces some bottlenecks ???

How long should it take to run the run_eval.sh script on an NVidia 1070 GTX GPU with 8GB ram ?
The ASUS motherboard has 24GB ram in it, and the hard drive is a 1TB SolidState.
Is this machine fit for the purpose of evaluating Alpha Fold ? Or are there any alternatives you could suggest ?

[Augustin-Zidek]

I am glad you managed to resolve most of your issues.

Wondering if this is because I'm not using a fast enough GPU or whether this Windows, Ubuntu, WSL setup introduces some bottlenecks ???

We tested AlphaFold only under Linux, so it is probable there is some performance issue when running it under WSL. It is even possible that TensorFlow can't use the GPU you have because it is running through WSL. I strongly recommend running AlphaFold on a Linux machine.

How long should it take to run the run_eval.sh script on an NVidia 1070 GTX GPU with 8GB ram ?

Which sequence are you running AlphaFold on? On a short sequence I would expect about 30-60 minutes. On a longer sequence a few hours.

Is this machine fit for the purpose of evaluating Alpha Fold ? Or are there any alternatives you could suggest ?

The machine specs look fine. As I said above, I would just recommend running AlphaFold under Linux and not through WSL as it can cause all kinds of weird performance issues.

[OneWorld-github]

Thank you for your feedback Augustin. I was just running the run_eval script as provided on github.

Checking NVidia Inspector I can see the GPU usage is indeed 0% at the following log stage when running the run_eval script

I0907 20:38:07.433789 140289582827328 contacts.py:114] Eval config is {'crop_shingle_x': 16, 'crop_shingle_y': 16, 'eval_sstable': 'alphafold_casp13/T1019s2/T1019s2.tfrec', 'max_num_exampl es': 500, 'output_path': '/home/user/contacts_T1019s2_2020_09_07_20_35_21/distogram/1', 'pyramid_weights': 0.75, 'save_rr_files': True, 'stats_file': 'alphafold_casp13/873731/stats_train_s 35.json'} num_bins: 64 I0907 20:38:08.370725 140737089963840 contacts.py:191] SepWorking on 0 T1019s2-l64_s24 T1019s2 64 I0907 20:38:08.394031 140737089963840 contacts.py:194] Getting residue_index from features
So your suspicions appear correct ! with WSL it doesn't seem to use the GPU !, at least with my current configuration.

I have a dual boot with Ubuntu installed. It's on a hard disc drive, so might be a bit slower than Windows which is running on a Solid State drive. Will give it a go, and report here how this goes also.